From: Artificial Intelligence in Clinics: Enhancing Cardiology Practice
| Type of test | Year | Authors | Brief summary | AI technology | Model input | Data source (Sample size) | Key findings | Reference |
|---|---|---|---|---|---|---|---|---|
| Medical interviews | 2020 | Harada Y. et al. | An LLM-based, automated medical history-taking system did not reduce waiting time for patients. | LLM | Papers, Journals, Guidelines, Electronic medical records, Public database, etc | Over 50,000 peer-reviewed medical articles, guidelines from the Japanese Society of Internal Medicine and the AMA, major medical journals, epidemiological data from CDC, WHO and others, etc | The system may improve the quality of care by supporting the optimization of staff assignments. | 11 |
| Diagnostic dialogue | 2024 | Tu T. et al. | Diagnostic accuracy of conversational medical LLM optimized for diagnostic dialogue was assessed as higher than that of primary care physicians. | LLM | Multiple-choice medical question answering, expert-curated long-form medical reasoning, electronic health record note summaries, and large-scale transcribed medical conversation interactions | 11,450 USMLE multiple-choice style open domain questions with four or five possible answers, 64 long-form medical question answering from MultiMedBench, 65 clinician-written summaries of medical notes from MIMIC-III, 89,027 audio transcripts of medical conversations during in-person clinical visits | This study does not have real-world patients. | 14 |
| Assessment of fraility | 2024 | Mizuguchi Y. et al. | Frailty assessment using ML models created from clinical information and features generated from walking videos by DL is associated with the risk of all-cause death in elderly patients with heart failure. | DL, ML | Walking video and clinical information | 417 patients with chronic heart failure over 75-year-old | Excellent agreements between the actual and predicted clinical frailty scale. | 17 |
| ECG and heart sound | 2023 | Shiraga T. et al. | ML models that integrate auscultation and ECG can efficiently detect conditions typically diagnosed via imaging. | ML | Raw PCG data, cropped ECG data, and echocardiography diagnosis | 1,052 patients undergoing echocardiography | Patients could be screened for severe AS, severe MR, and LVEF <40%. | 22 |
| ECG | 2010 | Kosmicki DL. et al. | The acoustic cardiographic model can predict LV systolic dysfunction. | ML | ECG and acoustic cardiographic data (S3, S4, and systolic time intervals) | 433 patients who had ECG, echocardiography, and BNP | This model outperformed BNP alone for predicting LV systolic dysfunction. | 23 |
| PCG | 2008 | Efstratiadis S. et al. | Assessed the correlation between systolic dysfunction and EMAT. | ML | ï¼…EMAT from PCG, findings of echocardiography, and left heart catheter data | 25 patients undergoing echocardiography, left-side heart catheterization, and PCG | An abnormal %EMAT was strongly associated with impaired LV dysfunction. | 24 |
| ECG | 2023 | Al-Zaiti S S. et al. | AI outperformed both precision and sensitivity in detecting NSTE-ACS. | ML | Raw ECG data | 7,313 patients with chest pain | AI helped correctly reclassify one in three patients. | 33 |
| ECG | 2019 | Attia Z. et al. | AI enabled identification of atrial fibrillation in ECG acquired during normal sinus rhythm. | CNN | Raw ECG data | 180,922 patients and 649,931 ECGs | AI identified atrial fibrillation with an AUC of 0.87. | 36 |
| ECG | 2021 | Yao X. et al. | The use of an AI algorithm based on ECGs can enable the early diagnosis of low EF. | CNN | Raw ECG data | 22,641 patients without a history of heart failure | More echocardiograms were obtained in the AI-positive ECGs. | 37 |
| ECG and echocardiogram | 2021 | Goto S. et al. | AI models with ECGs enhanced the performance of echocardiography models. | CNN | Raw ECG data and raw echo images | 5,495 studies for derivation, 2,247 studies for validation, and 3,191 studies for testing | Echocardiography model performance improved at 67% recall from PPV of 33% to PPV of 74-77%. | 38 |
| ECG | 2022 | Tison GH. et al. | AI-ECG can evaluate HCM status and treatment response. | CNN | Raw ECG data | 216 patients diagnosed with HCM | HCM scores by AI-ECG correlated with LV outflow tract gradients and NT-proBNP levels. | 39 |
| ECG | 2021 | Cohen-Shelly M. et al. | AI-ECG can identify patients with moderate or severe AS. | CNN | Raw ECG data | 258,607 patients undergoing echocardiography and ECG | The performance of the AI model increased with age and sex (AUC 0.90). | 41 |
| X-ray | 2024 | Bhave S. et al. | AI analysis of X-rays may be useful in the early identification of patients with LV hypertrophy or dilation. | DL | Chest X-ray images | 71,589 X-rays from 24,689 patients | The model outperformed all 15 individual radiologists in predicting LV hypertrophy or dilatation. | 45 |
| X-ray | 2023 | Saito Y. et al. | PAWP estimated from X-ray was useful for identifying and monitoring pulmonary congestion. | DL | Chest X-ray images | 534 patients admitted for acute heart failure | PAWP calculated by X-ray was significantly associated with higher event rates. | 8 |
| X-ray | 2021 | Homayounieh F. et al. | AI may improve diagnostic performance of radiologists in detecting pulmonary nodules on chest X-ray. | DL | Chest X-ray images | 100 X-rays | Junior radiologists saw greater improvement in sensitivity for nodule detection with AI compared with their senior counterparts. | 48 |
| X-ray | 2024 | Weiss J. et al. | AI may help identify individuals at high risk from X-ray when ASCVD risk score cannot be calculated. | DL | Chest X-ray images | 8,869 patients with unknown ASCVD risk score and 2,132 patients with known risk score | ASCVD risk of 7.5% or higher as predicted by AI had a higher 10-year risk for MACE after adjustment for risk factors. | 52 |
| Echocardiogram | 2021 | Narang A. et al. | AI allows novices without experience in ultrasonography to obtain diagnosis for evaluation of LV size, LV function, RV size, and pericardial effusion. | DL | Raw echo images | 240 patients examined by eight nurses | Nurse and sonographer scans were not significantly different for most parameters. | 55 |
| Echocardiogram | 2023 | He B. et al. | Initial assessments of LVEF by AI was noninferior to assessment by sonographers. | DL | Raw echo images | 3,769 exams | The AI saved time for both sonographers and cardiologists. Cardiologists were not able to distinguish between the AI and the sonographer. | 57 |
